Following is a list (in progress) of resources on Reinforcement Learning and allied topics. Though not comprehensive, it includes University lectures, YouTube playlists, MOOCs, blogs, etc. Hope this would be useful to someone getting started in reinforcement learning. UC Berkely CS 285 at UC Berkeley Deep Reinforcement Learning (23) Deep RL Bootcamp (15) CS 294: …
In this post, I am going to give a summary of nine key papers from the distributional reinforcement learning (DRL) area. Paper 001 : A Distributional Perspective on Reinforcement Learning This is the seminal paper in this area. The key idea of the paper is the argument that the value distribution is important in reinforcement …
Nine key papers in Distributional Reinforcement Learning Literature Read More »
Bellman operators come in Reinforcement Learning (RL). When I first encountered it, I had many questions regarding it. I often feel it is interesting to observe that what feels intriguing to someone. Many questions surface out in our minds. Why is it called an operator?. What are the inputs to it and output from it?. …
Overview In this note about distributional reinforcement learning, I am going to reflect on the paper titled A Distributional Perspective on Reinforcement Learning. I will try to give an overview of the underlying ideas behind this paper. Please keep in mind that a research contribution is often a culmination of multiple ideas/components. This makes from …
Summary of research paper “A Distributional Perspective on Reinforcement Learning” Read More »
Introduction In this tutorial, I am going to code the iterative policy evaluation algotithm from the book “Reinforcement Learning: An Introduction by Andrew Barto and Richard S. Sutton”. I am going to take psuedo code, image and examples from this text. The example I am taking for this tutorial is the gird world maze from Chapter …
Iterative Policy Evaluation for Estimating Value Function Read More »
Introduction Iterative Policy Improvement (IPI) is an algorithm in reinforcement learning to find the optimal course of action given the enviroment conditions. This blog post explains how it is done using a simple grid world navigating example. It works by iteratively improving an initial policy using the policy evaluation and policy improvement steps. Here’s how …
Introduction Bandit problems are the simplest possible reinforcement learning scenario. Here the bandit machine can have k arms and pulling each arm leaves the user a reward. One of the arms will be giving higher rewards in the long run and moreover this pattern could be changing over a time period. Think of the scenario …